Social responsibility at company level and inclusion of disabled persons: the case of Norway

This study investigates the relationship between social responsibility and disability policies in a sample of Norwegian enterprises, within the framework of sub-goal two of the agreement on a more inclusive working life (IA agreement). Incorporating elements of corporate social responsibility theory into a Nordic context, our main argument is that the presence of certain workplace arrange- ments and strategies associated with a social dimension increases the likelihood that employers will have implemented measures aimed at facilitating the inclusion of disabled people in the workplace. The results of the regression analysis generally confirm our expectations: having anchored work with the IA agreement within the general health, safety and environment work and having a regular contact with the Working Life Centres appear to be important organizational resources, significantly influencing the likelihood that enterprises make efforts to include disabled individuals.


Introduction
High labour force participation has long been a characteristic feature of the Norwegian labour market, and compared to other Organisation for Economic Cooperation and Development (OECD) countries Norway has a high level of employment; in the second quarter of 2011, 73.9% of the population between 15Á 66 years was employed (AKU 2011). Another positive characteristic of the Norwegian labour market is the participation rate of women, which is high in an international perspective (Report to the Storting No. 9 (2006Á2007)). However, if we look at the employment situation of individuals with different kinds of mental and physical disabilities, the picture is far less positive; of persons aged 15Á66 who reported having a disability, 42.3% were employed in the second quarter of 2011 (Bø and Håland 2011). While 47% of disabled individuals with employment had a parttime job in the second quarter of 2011, only 26% of the general working population were in part-time work (ibid.). Furthermore, far more persons with disabilities want to work than actually do work; in the second quarter of 2011, this applied to every fourth disabled person (ibid.).
Efforts made by employers to improve social inclusion are regarded as essential to reach the goals in the active social and labour market policy in Norway (see NOU 2000:27; Report to the Storting No. 9 (2006Á2007)), and according to Anvik et al. (2007), there is a need for workplace studies in the disability area to find out more about the influence of processes, relationships and dynamics in the workplace on the employment situation of people with disabilities. This article, which is based on a random sample of Norwegian IA enterprises combining both survey data and register data, aims to explain differences between employers regarding the extent to which they have implemented a policy aimed at facilitating the inclusion of disabled individuals. Our main hypothesis is that workplaces that display certain organizational characteristics and practices will be more inclined to actively seek to implement policies and strategies towards disabled people. Our analyses were conducted within the framework of sub-goal two of the agreement on a more inclusive working life (IA agreement), which highlights the need to bring disabled individuals into work. While drawing on the literature on corporate social responsibility (CSR), we have concentrated on the part of the CSR literature that is concerned with the responsibility of employers in 'societal questions', including participation in working life (see for instance Bredgaard 2004;Holt 1998).
The next section provides a more detailed description of sub-goal two of the IA agreement and presents relevant literature on CSR. Then our hypotheses are outlined, after which a brief description of data and methodology is provided. Finally, the results from the analyses are presented, whereupon these are summarized in a conclusion.

Social responsibility, disability and the IA agreement
The agreement on a more inclusive working life (IA agreement) was introduced in Norway in 2001. The agreement, as signed by the Norwegian government and the employer and employee organizations at the national level, committed the parties to make systematic efforts concerning the following three national goals: to reduce sick leave rates by at least 20%, bring more people with reduced work capacities into work and raise the real pension age (i.e. the average age of leaving working life). The agreement was renewed in 2005 and 2010.
The aim of sub-goal two of the IA agreement is to contribute to increased employment of people with reduced functional ability. However, in the course of the years the agreement has been in effect, it has become increasingly clear that it is in relation to this sub-goal that the major challenges in terms of implementation and achievements exist. In 2009, a comprehensive evaluation conducted by Ose et al. (2009) found that measures to reduce sickness absence were most prioritized (cf. Lie 2008). Thus, the version of the agreement signed in 2005 stated that a sharper focus on sub-goal two was needed. As a consequence, a supplement that described the aims for sub-goal two at national and organizational levels was issued in 2006, specifying the following target group: persons in work who either already suffer from or acquire reduced functional abilities and for whom employers have the main responsibility, and persons not in work who have reduced functional abilities and for whom public authorities, through social security schemes, have the main responsibility. Of particular importance to the latter group is preventing recipients of short-term benefits from becoming long-term recipients of social security benefits, as several studies have shown that working is 'healthy' for the individual, both physically and mentally. Clark (2003), for instance, found that unemployed had a poorer life quality than employed, and that mental health problems were more common among unemployed than employed. Moreover, the risk of getting sick was reduced, if one returned back to work (ibid.). Studies have also shown that those who become unemployed due to external causes, such as restructuring and downsizing, experience a higher risk of becoming ill than those who retain their work (Waddell and Burton 2006;Naess 2001). Naess (2001) also found a higher risk of suicide among individuals being unemployed, compared to individuals being employed. Another indication that participation in the workplace has a positive impact on the individual's health can be found in studies showing that job insecurity is associated with mental problems (see e.g. Lau and Knardahl 2008).
What, then, is meant by CSR? And to what extent can the Norwegian IA agreement be seen within such a theoretical framework? As a theoretical discipline CSR is first and foremost characterized by the fact that it is a wide-ranging field, with numerous definitions and interpretations (Aguilera et al. 2007;Dahlsrud 2008;Garriga and Mele 2004;Holt 1998;McWilliams, Siegel, and Wright 2006;Rosenstock et al. 2005;van Marrewijk 2003). However, in spite of the difficulty of reaching a clear definition of CSR, there is widespread consensus that CSR involves the social and environmental awareness and concerns of enterprises (see e.g. Aguilera et al. 2007;Dahlsrud 2008;Garriga and Mele 2004;Neal 2008). Windsor (2006) suggests that CSR can be regarded as any concept concerned with how employers handle social and public issues. What many definitions have in common is that CSR concerns what corporations actually are doing Á beyond the requirements of the law and beyond mere concerns of improving the company's financial results (see Carroll 1999;Rosenstock et al. 2005;Trygstad 2006).
In this article, we focus on the social dimension of CSR, which, briefly stated, concerns the willingness of enterprises to incorporate social concerns into their business activities and operations. Moreover, embedded in the part of CSR that is related to social issues, there is some form of social expectation that a legitimate corporation would act in a certain manner (Moir 2001), and according to Markel and Barclay (2009), increasing the employment of persons with disabilities should be regarded as a socially responsible initiative.
The relationship between CSR and the efforts of enterprises to include individuals with reduced work capacity has been paid little attention in the literature. For example, in a historical review of the concept of CSR conducted by Thomas and Nowak (2005), terms like 'disabled' or 'impaired' are not once mentioned. In a Nordic context, however, the association between companies' social engagement and efforts vis-à-vis disabled people is well known. Holt (2000) emphasises that the understanding of socially responsible businesses is very different in the USA and Denmark. Moreover, in Norway and elsewhere in Scandinavia, social responsibility is frequently associated with the willingness of companies to help solve problems of social inclusion, a willingness that can probably be attributed to the recognition that employment problems and social problems cannot be solved solely by the authorities and the welfare state (see Rosdahl 2000). Bredgaard (2004) argues that initiatives in Denmark in recent years to enhance the inclusion of different groups in working life have been founded on the idea of socially responsible companies. In Denmark, the term 'companies' social engagement' is widely used (see for instance Rosenstock et al. 2005), as a narrower concept than CSR. Social engagement focuses on what companies are doing voluntarily to ensure an inclusive labour market (Thuesen and Holt 2010).
Similarly, for 10 years, Norway has had the IA agreement, which is more or less explicitly based on the assumption that Norwegian businesses are willing to assume social responsibility for the inclusion of vulnerable groups in society. As with similar initiatives, volunteering is a keyword; apart from the Working Environment Act's general provisions on discrimination in the workplace, Norway has no statutory regulation of the social responsibility of companies. Hence, a precondition for socially responsible behaviour is that enterprises themselves are willing to integrate social concerns and responsibility into their everyday business models (see e.g . Boll 2002;Rosdahl 2000;Thuesen and Holt 2010).
Traditionally, CSR has not played an important role in countries with a social democratic welfare model as in Norway, primarily because social responsibility, mostly for political reasons, has been placed almost exclusively on the state (Boll 2002). However, with the introduction of the IA agreement in 2001, the role of social responsibility increasingly came into focus in Norway. The combined focus on increasing the supply of labour and the inclusion of groups experiencing difficulties in entering the labour market led Olsen, Svendal, and Amundsen (2005) to define the IA agreement as a programme being placed within 'the workfare philosophy'. Since the IA agreement was first signed in 2001, demands on the employers' contribution in terms of adjustment and supervision for their employees have increased. However, and as we discuss in more detail in the next section, the scope for acting in a socially responsible manner will be influenced by the presence of certain workplace practices and structures.

Hypotheses
Our main hypothesis was that workplaces that exhibit practices associated with a social dimension will be more likely to pursue active policies vis-à-vis disabled individuals than enterprises that have not developed such practices. More precisely, we argued that enterprises that (i) have anchored work in relation to the IA agreement within their general health, safety and environment (HSE) work, (ii) have signed the agreement early in the period (2001 or thereabouts), (iii) are in regular contact with local unions and (iv) are in regular contact with the Working Life Centre would be at an advantage in implementing and realizing sub-goal two Á in terms of interests, familiarity, culture and the organizational framework in which they operate. Moreover, what these practices have in common is that they are aimed at providing better working conditions, strengthening cooperation and increasing inclusion at the workplace. As such, they reflect the social responsiveness of organizations. Rather like Martin (2005), who found that differences in workplace organization and practices could explain cross-national variations in employers' attitudes to social policy programmes, and Thuesen and Holt (2010), who found that strong social relationships and cooperation in the workplace had a positive effect in terms of lowering employee turnover, we assumed that differences in these workplace factors would have an effect on whether enterprises work actively to realize sub-goal two of the IA agreement.
Our first hypothesis was based on the assumption that enterprises that incorporate work related to the IA agreement into their general HSE efforts will have an organizational advantage when it comes to implementing sub-goal two because efforts and initiatives then become a part of organizational practices that are official and legitimate in the organization. According to the Working Environment Act, HSE is considered to be a work environment measure. Moreover, the existence of a HSE system itself can be interpreted as an indication that a business makes serious efforts to establish socially responsible organizational practices, by obeying the law and being committed to ensuring a good working environment (Campbell 2007). Amick et al. (2000) found that a company's safety climate, considered an organizational policy and practice, did predict the safety and prevention activities in which an organization invested in order to reduce incidence of work disability, while Martin (2005) found that the degree of professionalization of human resource management (HRM) activities in a business affected positively its participation in social programmes. Markel and Barclay (2009) point out that HRM activities, as an area of the organization, represent a 'resource in its disability supportive practices and expertise'. According to Bredgaard (2004), the existence of work environment measures is an important indicator that the enterprise maintains internal social responsibility. Hence, our first hypothesis was that companies that incorporate IA efforts into their HSE practices will be more active in pursuing sub-goal two of the IA agreement.
Our second hypothesis concerned the number of years that have passed since signing the IA agreement. According to Trygstad (2006), the adherence of a company to the IA agreement can be regarded as an indicator of its degree of social responsibility. We took this argument further, by arguing that the number of years since a company signed the agreement will influence the likelihood that measures and strategies to facilitate the inclusion of disabled persons have been initiated. Our basic assumption was that companies that joined the agreement early on had signalled a greater willingness to work with issues related to sickness leave, inclusion and prevention of exclusion, and thereby a preference to act in a socially responsible way. Moreover, businesses that signed the agreement early in the period were willing to accept that both the costs and the benefits of signing the agreement were unknown. A reasonable interpretation would thus be that these early signers were genuinely motivated by a desire to help solve social problems related to inclusion and participation, in spite of the uncertainties and risks. Our position here was also supported empirically by Dale-Olsen, Hardoy, Storvik, and Torp (2005), who found that the number of new appointments and cases of retaining employees with disabilities was highest in companies that had signed the IA agreement early in the period (2001Á2002). Anvik et al. (2007) argued that the longer a company has had the IA agreement, the greater the probability that it has employees with disabilities or that it has recently appointed such employees. One explanation for this could be that since the agreement is also about influencing attitudes, the longer the agreement has been effective, the higher is the likelihood that attitudes actually are changed. Hence, our second hypothesis stated that the longer the period with IA agreement, the higher the probability that a company will be actively implementing sub-goal two of the agreement.
Local labour unions often act as important drivers in matters concerning social aspects of working life, such as for example work environment, workplace conditions and social inclusion (Campbell 2007;Rosdahl 2001;Rosenstock et al. 2005). According to Cramer (2003), unions, through collective bargaining, could influence the social policies of companies and 'create room for specific target groups in the labour market'. Trygstad (2006) argued that collaboration in itself may be regarded as an indicator of social responsibility at company level. Likewise, Campbell (2007) argued that industries demonstrate social responsibility by establishing regulatory mechanisms to ensure fair practices and workplace safety. We adopted theoretical interpretations of this sort here; namely, that companies that cooperate with trade unions demonstrate a willingness to achieve dialogue and cooperation on important workplace issues. Our reasoning was similar to that of Amick et al. (2000), who took both the willingness of companies to involve employees in plans and decisions and the degree to which working relationships were cooperative as indicators of the organizational policies and practices of organizations.
Unions could act as driving forces, pushing employers to follow-up their obligations under the IA agreement. Furthermore, enterprises that have established a tradition of collaboration in related areas could more easily integrate work related to the IA agreement within existing arenas of collaboration. Our third hypothesis was thus that businesses that are in regular contact with their local union(s) have gone further in realising sub-goal two.
The Working Life Centres are resource and competence centres for enterprises that have signed the IA agreement, and signing the agreement gives organizations access to a contact person in the local centre. The contact persons support organizations in their IA efforts, and they are among the most often used measures within the IA agreement (Ose et al. 2009). Ose et al. (2009) also found that some employers emphasized that the Working Life Centres were important sources of knowledge and information in their efforts to realize the sub-goals of the IA agreement (see also Dale-Olsen, Hardoy, Storvik, and Torp 2005). The centres are intended to be a source of knowledge and to act as a push factor to ensure continuity in IA efforts at the enterprise level. By extension, the regularity and stability of the dialogue that enterprises have established with their local Working Life Centre will influence how far they have come with regard to the implementation of sub-goal two. Our expectation, as stated in hypothesis four, was that employers that have regular contact with the Working Life Centre will be more likely to work actively to realise sub-goal two of the IA agreement.

Data and method
This article merged survey data and register data. The management in 5000 IA companies that replicated the industrial mix of the Norwegian IA register received a questionnaire in 2008. The composition of enterprises in the sample reflected the composition of enterprises in the register, with regard to sector and industry (on threedigit industry code level, i.e. random selection within 164 industry codes). The 5000 enterprises that were sampled represented 20% of the enterprises in the IA register at the time of the survey in 2008. Random selection further ensured that the sample was representative in terms of number of employees. No enterprise in the sample had less than nine employees. The rationale behind this decision was that enterprises with fewer than nine employees, based on qualitative exploratory interviews that were conducted prior to the survey, had little information to share concerning their internal IA work. Moreover, the vast majority of small businesses are in the private sector and the private sector has a lower proportion of IA businesses than the public sector. Finally, organization number was used to identify units in the survey, which ensured that we reached out to the organizational level where the concrete experiences with the IA work are found. Once the cross-sectional survey data had been gathered, they were linked with register data (number of employees, sickness absence rates, sector and industry) derived from the Norwegian Sick Leave Register and the Business Register. Statistics Norway merged the data for us.
The results presented here are based on the responses from the managers, of whom 50% answered the questionnaire. Adjusted for units missing value on one or more of the variables in the model (non-response), we ended up with N 01836 (net sample in the first regression analysis). This corresponds to a response rate of 36%. Table 1 compares the distribution among businesses in the IA register (based on our representative gross sample) and our net sample on the industry variable. The last column shows the difference (%) between them within each industry. Table 1 shows a high degree of correspondence between the enterprises in the IA register and enterprises in our net sample with respect to industry. For 11 of the 15 industries, the variance is within 1% (plus or minus). The net sample is thus considered to be representative of the businesses in the IA register. We also checked the correspondence between the companies that responded to the survey (N 02425) and those that constituted the net sample (N 01836). The comparison revealed that the variance between the two samples was minimal. Moreover, observations missing in the analyses appeared to be relatively randomly distributed across industry.

Dependent variables
The social responsibility of enterprises typically focuses on three types of behaviour (Rosdahl 2001;Rosenstock et al. 2005): (1) Retention in work (e.g. of long-term sick people, other employed persons with reduced working ability or elderly people); (2) Prevention (e.g. of health problems and social problems among employees) and (3) Integration (i.e. the hiring of for instance long-term unemployed people or persons with reduced working capacity). Following the distinction made by Holt (1998) and Bredgaard (2004), the first and second concern companies' internal social responsibility while the third concerns their external social responsibility. In our analyses, we considered variation in both internal and external social responsibility; in the first two regressions, internal responsibility was assessed by applying the following dependent variables: The enterprise adapts working conditions for employees who have disabilities and the enterprise has set activity goals for the follow-up and integration of employees with reduced working ability. The third regression concerned external social responsibility, measured by the following variable: The enterprise makes efforts to recruit persons who have permanent disabilities. Common to the three dependent variables is that they can be seen as performance variables, assessing the efforts of enterprises towards people with disabilities.
We would like to emphasize that the use of the term permanent in the analysis of external social responsibility was justified by a desire to make a clear delineation of the target group. Moreover, we wanted to 'push' the respondents on the question of how far they actually feel that their responsibility extends, when it comes to the recruitment and inclusion of individuals with permanent disabilities, who have no relation to their business or the labour market in general.
Since all three variants of the dependent variable originally had few values, we decided to reduce the number of values to 2, and used binary logistic regression to test the hypotheses. Briefly, this technique implies that we can predict the probability of occurrence of a certain event on the dependent variable (Logit L) by fitting data to a logit function logistic curve. Explanatory variables (covariates) are included in the model, and through maximum likelihood estimation, the values of the independent variables that are most likely to have defined the values on the dependent variable in the sample are estimated.

Independent variables
On the right-hand side of the regression equation, the variables measuring contact with the local union and contact with the Working Life Centre initially had the following values; 'never', '1Á4 times per year', '5Á10 times per year', 'approximately once a month', 'more often than once a month', 'about once a week' and 'daily contact'. Since these variables were categorical (ordinal level), we had to recode the variables into dummy variables. While the first four values were retained, the last three values were merged into one dummy variable (more than once a month/weekly/ daily). The variable measuring whether work in relation to the IA agreement is anchored in the company's HSE efforts originated from a question that had the following response options: 'totally disagree' (1), 'partially disagree' (2), 'neither or' (3), 'partly agree' (4) and 'strongly agree' (5). The attitude variable (the company cannot hire people who are unable to provide 100%) initially had the same values. Both these category variables were recoded into dummy variables with identical values; 'do not agree' (based on initial value 1Á3), 'partly agree' (initial value 4) and 'fully agree' (initial value 5).
The sector variable was dichotomous, with value 1 (public sector) and 0 (private sector). It should be noted that the proportion of public enterprises in the IA register is far higher than in the Norwegian economy in general; in 2012 public enterprises constituted 53.4% of the enterprises in the register (state and local enterprises), compared to 14% in the economy in general. Business size was coded into the following four dummy variables; '9Á19 employees', '20Á49 employees', '50Á99 employees' and '100 employees and more'. Industry was recoded into six dummy variables, where industries sharing essential similarities (in terms of production, qualifications and educational requirements) were merged with each other. Information on sickness absence level (doctor-certified sickness absence rate, company average level from 2008) was included as a continuous variable. Finally, number of years with the IA agreement ranged from 1 year to 8 years. Table 2 presents descriptive statistics for the variables and observations included in the regression analysis.
The decision to test the same explanatory model in all the analyses was both theoretically and empirically motivated. Theoretically, it makes sense to see internal and external social responsibility as somewhat related; if a business is working to facilitate the work situation of employees with reduced working capacity, then it is likely that this business is more aware of the importance of recruiting people with disabilities from the outside, compared to a business that does little to take care of its own employees in terms of adaptation and facilitation. Both kinds of responsibility (internal and external) are part of the IA agreement, and central to the agreement is the idea that the various measures that are made available to the IA enterprises (such as assistance via Working Life Centres) should impact positively on the realization of both the internal and the external 'sides' of sub-goal two. The empirical motivation behind conducting the analyses in the manner we have done was simply to examine the explanatory power of our theoretical model in light of different dependent variables, focusing on both internal and external social responsibility. Are the explanatory variables most important in explaining variations in internal or external social responsibility? This should be of both theoretical and practical interest.
Upcoming tables report coefficients, standard errors of the estimates and pvalues. Goodness-of-fit is reported in terms of Nagelkerke R 2 , although measures of model fit in logistic regression should be regarded as less reliable than in linear regression analysis (see for instance Hosmer and Lemeshow 2000). Analyses were conducted in the sample as a whole and among private and public sector enterprises separately. However, limited empirical variance, in combination with small sample size, forced us to drop the industry dummy variables from the sector-based analysis. Likewise, the analysis of external social responsibility within the private sector had to be dropped, due to small sample size and a skewed empirical distribution on the dependent variable. The statistical analyses were carried out using IBM SPSS Statistical software version 19.

Results
Explaining differences in the efforts of enterprises with regard to sub-goal two: internal efforts vis-a`-vis disabled persons Tables 3 and 4 go far in supporting our overall hypothesis that certain workplace arrangements will have a positive impact on the efforts made by enterprises concerning the implementation of sub-goal two. Integrating work related to the IA agreement within a company's HSE programme and having a regular contact with the Working Life Centre are both significantly associated with the likelihood that employers measure positively on the two dependent variables that involve internal social responsibility. The results are particularly strong and significant for the dummy variables measuring the extent to which work in relation to the IA agreement is anchored in the HSE work, but also for the dummies measuring the degree of contact with the Working Life Centre, several positive and significant associations appear. However, the 'relative strength' of the dummies in some cases differ slightly from our theoretical expectation, for instance when the association between having contact with the Working Life Centre 5Á10 times a year and the dependent variable in Table 3 is more pronounced than is the case with the dummies that measure a more frequent contact. The other two workplace variables we expected to be important in relation to efforts vis-à-vis people with disabilities, namely number of years with the IA agreement and degree of contact with the local union, turned out to have little impact Á the exceptions being in Table 3, where a significant difference appears within the private sector (as we move from 'no contact' (reference category) to '1Á4 times per year') and in Table 4, where having contact with local unions more than once a month/weekly/daily increases the likelihood that an enterprise has set activity goals for the follow-up and integration of employees with disabilities, when compared to the reference category (no contact) and when all other controls are held constant. In addition, the coefficient for number of years with the IA agreement is positive and significant in the same table, within the public sector.
As expected, companies that fully agreed with the general statement that the company can hire people who are unable to provide 100% (i.e. positive attitude) are more likely to adapt the workplace for disabled employees than businesses that reveal a negative attitude. In contrast, attitudes are not associated with the other dependent variable measuring internal effort (the enterprise has set activity goals for follow-up and integration of employees with reduced working ability). A possible explanation to this is that setting activity goals is a more narrow activity than adjusting work conditions, and thus is less likely to capture the actual variation in internal effort.
Turning now to level of sickness absence, this variable is neither positively nor negatively associated with the internal efforts of enterprises. Hence, an expectation that a low sickness level may serve as an expression of an enterprise focusing on absence prevention and inclusion in the workplace, and thus could be positive with respect to sub-goal two, is not supported. What, then, about structural variables like company size, sector and industry? In none of the regressions did these variables turn out to have a significant impact on the two dependent variables that measure the degree of internal commitment to sub-goal two Á with two exceptions; being a public sector enterprise increases the probability that the enterprise will adapt working conditions for employees with disabilities, and being an enterprise in the health and social services, other social services and personal services does the same, when compared to enterprises in agriculture, forestry and fishing, mining and quarrying, manufacturing, electricity and water supply, building and construction. These findings are consistent with those of Holt (1998), who found that the public sector tends to be more socially responsible than the private sector (see also Rosenstock et al. 2005). Similarly, Thuesen and Holt (2010) found that the public sector is more open than the private sector, and that there are differences between the industries concerned, i.e. that industries such as agriculture and infrastructure are less inclusive than for instance health and welfare. An explanation, according to Thuesen and Holt (2010), is that industries such as agriculture and infrastructure involve demanding physical work, which means that they have less room for people with reduced work capacity. However, it should be pointed out that the use of a sector variable in combination with the industry dummies may have reduced the explanatory power of industry in our analyses, since the division into different industries in our data largely coincide with the division between public and private sector (85% of the enterprises within health and social services in our data are public, while this applies to only 3% of the enterprises in the reference category).

Explaining differences in enterprises' efforts with regard to sub-goal two: external efforts vis-a`-vis disabled persons
The question of whether enterprises make efforts to recruit persons with permanent disabilities was used to measure external social responsibility. Our main hypothesis that the presence of certain workplace practices associated with a social dimension will positively influence efforts to include disabled individuals in the workplace is to a substantial extent confirmed (Table 5). Again, having integrated work related to the IA agreement into the HSE efforts, and being in regular contact with the Working Life Centre are factors that are positively associated with the likelihood that employers work actively to recruit persons with permanent disabilities. On the other hand, we see that most of the dummies measuring contact with local labour unions seem to be negatively associated with the likelihood that a company attempts to recruit persons with disabilities. Although only one of the coefficients actually is   (2007), who found that companies with high levels of union membership actually had proportionately fewer disabled employees than other companies. Our result may also reflect that sub-goal two is partially targeted at a group of disabled individuals who are outside the labour market, and hence also are unorganized. Thus, unions might not regard it as their responsibility to influence the situation of a group of individuals who are not their members. Moreover, there is a possibility that enterprises, 'in agreement' with the unions, are more concerned with implementing preventive measures for their employees than in recruiting people from outside (see Holt 1998). Nor should we forget that the presence of unions within enterprises matters only if the unions actually urge the enterprises to move in a more socially responsible direction (Martin 2005). The expectation that the number of years with the IA agreement would influence employers' efforts to include disabled individuals, finds no support in the analysis. The results probably indicate that having signed the agreement is by no means sufficient by itself. The results further suggest that the agreement in some enterprises is a 'sleeping agreement', which may be particularly true for enterprises that signed the agreement early in the period (2001Á2002). In any case, nothing in our findings indicate that companies that signed the agreement early in the 'IA era' are more socially responsible (internally or externally) than those enterprises that have signed the agreement more recently.
Why, then, did some enterprises sign the agreement in the beginning, when the agreement still was new and uncertain, and both potential gains and costs unknown? The reason could be that the IA agreement, despite uncertainty about potential costs and benefits, still meant that the enterprises got access to some economic measures. A decision to sign the agreement early in the period might also reflect workplaces experiencing significant challenges with regard to sickness absence and exclusion of employees. In this light, joining the agreement early might simply reflect a sincere desire to achieve a more inclusive workplace. Nor should we rule out the possibility that becoming an IA enterprise for many businesses have been motivated by the opportunity to entitle themselves with the 'official' status of being an inclusive workplace, and thereby send a signal to the outside world that one is a socially responsible company. Turning now to the attitude variable (the company cannot hire people who are unable to offer 100%), our findings in Table 5 (work to recruit persons who have permanent disabilities), as in Table 3 (do adaptations for employees who have disabilities), are in line with what we expected; the more positive the attitude expressed, the higher the likelihood that enterprises make workplace adaptations for employees with disabilities and try to recruit persons with permanent disabilities. In Table 5, this applies to companies in the public sector as well as in the sample as a whole. The fact that both the attitude variable and the dependent variable in Table 5 focuses on companies' external efforts to increase the inclusion of disabled persons, have probably played a role here.
However, some methodological considerations should be mentioned with regard to the attitude variable. First, it is unlikely that the influence goes in only one direction. Effects could obviously go either way, in the sense that the absence of specific activities and strategies for people with disabilities may help to uphold negative attitudes. Attitudes do not exist in a vacuum. Second, it is problematic that only managers answered the questionnaire on behalf of the company; we cannot conclude that the business as such is characterized by positive or negative attitudes (cf. Holt 1998). Third, it may be questioned how suitable the variable actually is, as a measure of attitudes towards disabled in working life. It is quite easy to imagine that many small businesses have responded affirmatively on this question, simply because they consider real adaptation opportunities as minimal. Hence, answers may reflect practical realities at the workplace as much as attitudes to hiring disabled persons.
None of the coefficients for sickness absence rates are significant in our analysis. Our lack of expectations in advance was simply due to the recognition that the interpretation of potential relationships between sick leave rates and efforts and initiatives vis-à-vis people with disabilities is ambiguous. On the one hand, it would make sense to imagine that a low level of sick leave would positively affect efforts made in relation to sub-goal two, based on the interpretation that a stable, low level of sick leave quite often will be the result of targeted efforts to reduce absenteeism in the workplace (cf. Bredgaard 2004). In turn, we might expect that these general efforts to retain and include employees would make these enterprises more likely to incorporate strategies aimed at hiring and assisting disabled persons. On the other hand, we might easily imagine that low sickness absence reduces the likelihood that the company is actively making efforts in favour of people with disabilities because the 'quota of jobs' which can be accommodated in the enterprise is already used up by employees on sick leave. There is also the methodological problem that sickness absence levels probably are largely exogenously determined, i.e. they can be explained by factors that lie outside our specific model. Our results did little to clarify this theoretical and methodological ambiguity.
Turning now to the variable measuring company size, being an enterprise with 100 employees or more reduces the likelihood that the enterprise works to recruit individuals with permanent disabilities, when compared to the reference category (9Á19 employees) and when all other variables are held constant. However, the variable is characterized by a general absence of significant estimates in our analyses. In light of previous research that has found that large companies have better opportunities to adapt working conditions than small businesses, this is surprising (see for instance Holt 1998). Others, such as Martin (2005), have pointed out that business size might enhance support for social policy because large businesses are more likely to have experience with training programmes. Our analyses show something else, namely that what employers do, in terms of the weight they place on strategies and practices intended to improve cooperation, inclusion and work conditions, is more important than the sheer size of the enterprise. Moreover, the support for the idea that company size matters is practically inexistent in our analyses of variations in internal and external social responsibility. Model fit (Nagelkerke R 2 ) ranges from 0.094 to 0.148 in the three tables, indicating that there is much unexplained variance in our model. However, we must remember that measuring model fit in logistic regression is burdened with methodological challenges, since the logit is dichotomous. This in contrast to ordinary least squares (OLS) regression, where many values on the dependent variable ensure that the explained variance will be less sensitive to the gap between observed and predicted values. Moreover, explained variance is generally low in logistic regression (Hosmer and Lemeshow 2000).

Conclusion
Few attempts have previously been made to establish a theoretical link between organizational factors associated with social responsibility and the presence of disability policies and strategies at the enterprise level. Inspired by the literature on CSR, this article pursued a theoretical argument stating that certain cooperative workplace practices could be positive to the efforts made by enterprises with regard to retaining and including disabled individuals. Efforts vis-à-vis disabled persons were measured in terms of the implementation of sub-goal two of the IA agreement, which focuses both on the internal and external responsibility of enterprises in realizing a more inclusive working life. Overall, the analyses provided substantial support for our general expectation that the depth of both internal and external efforts vis-à-vis disabled persons is linked to the existence of certain workplace factors. The analyses first and foremost provided support for our hypothesis that integrating efforts in relation to the IA agreement into the HSE work in general will be positive with respect to employers' efforts towards disabled. This finding has important implications; enterprises that have established HSE routines will have an organizational framework that is also positive to the implementation of sub-goal two of the IA agreement. Moreover, linking work in connection with sub-goal two with the arenas and venues for cooperation that have already been established through the HSE work could provide both greater efficiency and legitimacy to the strategies and processes aimed at implementing sub-goal two. Based on the sector-specific analyses, however, it should be pointed out that our theoretical model seems to be slightly more relevant for the public than the private sector.
Considering now the Working Life Centres, our results were consistent with findings we referred to earlier in the article, namely that the relation with the Working Life Centre is the relation with which employers most often express satisfaction (see Dale Olsen et al. 2005;Ose et al. 2009). The results may reflect the fact that the Working Life Centres are regarded as important sources of information and knowledge. However, we cannot exclude the possibility that those enterprises that use the Working Life Centre are also those with most knowledge of the types of assistance they provide and of the IA agreement in general. As such, there is a possibility that companies that claimed to have benefited most from the collaboration were also those that were already best prepared in terms of knowledge and motivation. Moreover, based on our cross-sectional data, we cannot make causal claims that having anchored work in relation to the IA agreement within the HSE work, or having a regular contact with the Working Life Centre, is what makes employers do adaptations for disabled workers. Even though our results clearly point in such a direction, future research involving panel data is needed, in order to explore causal mechanisms more closely.
Some other methodological issues should also be mentioned. A limitation to the study is that the survey data only reveal what the managers perceive about their organizations. The quality of the data will thus be vulnerable to the managers' knowledge, understanding and general 'closeness' to the questions addressed. Furthermore, these are factors that are likely to vary between managers. In addition, (disabled) employees may have experiences and perceptions of the same questions which differ from the view of the leaders. Further, the indicators of social commitment and effort developed in this article by no means capture the full range of policies and strategies designed to strengthen the situation of disabled individuals in the workplace. Moreover, one can easily imagine other ways of measuring the social engagement and commitment of employers (cf. Holt 1998).
Another challenge to our theoretical model is that the employer behaviour that we describe in positive terms may be the result of considerations and assessments that originally had little to do with ambitions to act in a socially responsible way. Moreover, what apparently might be conceived as socially responsible behaviour among employers may not necessarily be motivated by a noble desire to become socially responsible (Rosenstock et al. 2005). There is also a risk that the pursuit of social responsibility can turn into a 'social ritual' that is more about the need to demonstrate a willingness to act and give the impression of action, rather than solving real problems (Bredgaard 2004). However, as a description of the Norwegian IA agreement, this 'image' is probably not the most adequate; since joining the agreement is voluntary, it is likely that those enterprises that have entered into the IA agreement in fact are motivated to make a contribution to increase participation and inclusion in the labour market.
Overall, our findings support the idea that the behaviour and actions of enterprises might be more important for their efforts vis-à-vis people with disabilities than factors that are more or less given in advance, such as sector, industry and company size. As such, the findings should give positive inspiration to further work on the IA agreement in Norway and to similar initiatives in other countries, in the sense that company behaviour matters. Moreover, incorporating these strategies within general HSE efforts and established collaborative practices would probably be beneficial to the wider inclusion of disabled persons in working life.